Face detection method and apparatus using level-set method

Abstract

A face detection method and apparatus using a level-set method. The method includes: laying a first initial shape model over a face image; evolving shape curves of the first initial shape model using the level-set method; detecting first through M-th neighboring boundary points that neighbor first through M-th initial boundary points of the first initial shape model, and that neighbor the evolved shape curves, M being a positive integer larger than 1; laying a modified shape model rendered using the detected first through M-th neighboring boundary points over the face image; detecting a rotation angle of the modified shape model; determining whether the detected rotation angle of the modified shape model is within a predetermined angle range; and selecting one of a second initial shape model, which is rotated to the left, or a third initial shape model, which is rotated to the right, according to the detected rotation angle, and laying the selected initial shape model over the face image, and returning to the detecting of the first through M-th neighboring boundary points, when the detected rotation angle of the modified shape model is outside the predetermined angle range.

1. A face detection method using a level-set method, comprising:
 laying, under control of a processor, a first initial shape model, which has a front viewpoint, over a face image; 
 evolving shape curves of the first initial shape model laid over the face image using the level-set method; 
 detecting first through M-th neighboring boundary points that neighbor first through M-th initial boundary points of the first initial shape model, and that neighbor the evolved shape curves, M being a positive integer larger than 1; 
 laying a modified shape model rendered using the detected first through M-th neighboring boundary points over the face image; 
 detecting a rotation angle of the modified shape model; 
 determining whether the detected rotation angle of the modified shape model is within a predetermined angle range; and 
 selecting one of a second initial shape model, which is rotated to the left, or a third initial shape model, which is rotated to the right, according to the detected rotation angle, and laying the selected initial shape model over the face image, and returning to the detecting the first through M-th neighboring boundary points, when the detected rotation angle of the modified shape model is outside the predetermined angle range. 
 
   
   
     2. The face detection method of  claim 1 , wherein, in the evolving shape curves using the level-set method, the shape curves are evolved using a set of points determined based on geometric information and the rotation angle of the first initial shape model. 
   
   
     3. The face detection method of  claim 1 , wherein the detecting of the first through M-th neighboring boundary points comprises:
 detecting a K-th neighboring boundary point that neighbors a K-th initial boundary point of the first initial shape model, K being a positive integer between 1 and M; 
 determining whether the detected K-th neighboring boundary point is within an evolution range of the shape curves; 
 setting the K-th initial boundary point, instead of the detected K-th neighboring boundary point, as a new K-th neighboring boundary point, wherein if the detected K-th neighboring boundary point is not within the evolution range of the shape curves; and 
 determining whether all of the first through M-th neighboring boundary points have been detected; 
 wherein, when the detected K-th neighboring boundary point is within the evolution range of the shape curves, the face detection method proceeds to the determining whether all of the first through M-th neighboring boundary points have been detected, and when not all of the first through M-th neighboring boundary points have been detected, the face detection method returns to the detecting the K-th neighboring boundary point. 
 
   
   
     4. The face detection method of  claim 1 , wherein, in the detecting of the rotation angle of the modified shape model, the rotation angle of the modified shape model is detected using coordinates of a portion of the modified shape model representing a mouth in the face image, coordinates of a portion of the modified shape model representing a center of one of plural eyes in the face image, and the distance between the eyes of the face image. 
   
   
     5. The face detecting method of  claim 4 , wherein the rotation angle of the modified shape model is detected using the following equation:
     A =tan −1 ((2×( d −2×( m−i ))/( d ×(5 −s )), 
 wherein A represents the rotation angle of the modified shape model, d is a distance between eyes of a person, m is the coordinates of a portion of the modified shape model representing the center of the mouth in the face image, i represents the coordinates of a portion of the modified shape model representing a left eye or a right eye in the face image, and (m−i) is half of the distance between the left and right eyes in the face image calculated using the distance between the center of the mouth in the face image and the center of the left eye or right eye in the face image. 
 
   
   
     6. The face detecting method of  claim 1 , wherein the predetermined angle is between −5 degrees and 5 degrees. 
   
   
     7. A face detection apparatus using a level-set method, comprising:
 a shape curve evolution control unit, which evolves shape curves of a first initial shape model laid over a face image using the level-set method, the first initial shape model having a front viewpoint; 
 a neighboring boundary point detection control unit, which detects first through M-th neighboring boundary points that neighbor first through M-th initial boundary points of the first initial shape model, M being a positive integer larger than 1; 
 a model rotation angle detection unit, which detects a rotation angle of a modified shape model rendered using the detected first through M-th neighboring boundary points; 
 an angle range determination unit, which determines whether the detected rotation angle of the modified shape model is within a predetermined angle range; and 
 a shape model arrangement control unit, under control of a processor, which lays one of the first initial shape model, the modified shape model, a second initial shape model rotated to the left, and a third initial shape model rotated to the right over a face image. 
 
   
   
     8. The face detection apparatus of  claim 7 , wherein the shape curve evolution control unit evolves the shape curves using a set of points determined based on geometric information and the rotation angle of the first initial shape model. 
   
   
     9. The face detection apparatus of  claim 7 , wherein the neighboring boundary point detection control unit comprises:
 a neighboring boundary point detector, which detects a K-th neighboring boundary point that neighbors a K-th initial boundary point of the first initial shape model, K being a positive integer between 1 and M; 
 a neighboring boundary point location determiner, which determines whether the detected K-th neighboring boundary point is within an evolution range of the shape curves; 
 a neighboring boundary point setter, which sets the K-th initial boundary point, instead of the detected K-th neighboring boundary point, as a new K-th neighboring boundary point; and 
 a neighboring boundary point detection determiner, which determines whether all of the first through M-th neighboring boundary points have been detected. 
 
   
   
     10. The face detection apparatus of  claim 7 , wherein the model rotation angle detection unit detects the rotation angle of the modified shape model using coordinates of a portion of the modified shape model representing a mouth in the face image, coordinates of a portion of the modified shape model representing a center of one of plural eyes in the face image, and a distance between the eyes of the face image. 
   
   
     11. A computer-readable storage medium encoded with processing instructions for causing a processor to execute a face detection method, the method comprising:
 laying a first initial shape model, which has a front viewpoint, over a face image; 
 evolving shape curves of the first initial shape model using the level-set method; 
 detecting first through M-th neighboring boundary points that neighbor first through M-th initial boundary points of the first initial shape model, and that neighbor the evolved shape curves, M being a positive integer larger than 1; 
 laying a modified shape model rendered using the detected first through M-th neighboring boundary points over the face image; 
 detecting a rotation angle of the modified shape model; 
 determining whether the detected rotation angle of the modified shape model is within a predetermined angle range; and 
 selecting one of a second initial shape model, which is rotated to the left, or a third initial shape model, which is rotated to the right, according to the detected rotation angle, and laying the selected initial shape model over the face image, and returning to the detecting the first through M-th neighboring boundary points, when the detected rotation angle of the modified shape model is outside the predetermined angle range. 
 
   
   
     12. A face detection method, comprising:
 laying, under control of a processor, a first initial shape model over a face image; 
 evolving shape curves of the first initial shape model laid over the face image using a level-set method; 
 detecting first through M-th neighboring boundary points, the neighboring boundary points neighboring first through M-th initial boundary points of the first initial shape model and the evolved shape curves, M being a positive integer larger than 1; 
 laying over the face image a modified shape model rendered using the detected first through M-th neighboring boundary points; 
 detecting a rotation angle of the modified shape model; 
 determining whether the detected rotation angle is within a predetermined angle range; and 
 selecting a second initial shape model or a third initial shape model according to the detected rotation angle, and laying over the face image the selected initial shape model, when the detected rotation angle is not within the predetermined angle range.